[0001] This invention relates to a bias voltage generator for static CMOS circuits on a
Very Large Scale Integration (VLSI) chip.
[0002] Most VLSI chips are now being designed with CMOS technology. It is believed that
this trend is an attempt to capitalise on the improved power and performance characteristics
which the CMOS technology provides. A typical VLSI chip is made up of thousands of
circuit arrangements integrated on a single substrate. Some of these circuits, such
as programmable logic arrays (PLA) and busses with several dotted drivers, require
static power dissipating circuits in order to ensure proper operation.
[0003] One of the disadvantages of these static circuits is that, at worst case, they dissipate
a relatively large amount of power. Because CMOS technology is a lower power technology,
the presence of static circuits on the CMOS chip negates or undermines one of the
benefits associated with the CMOS technology. Even though there are a plurality of
different types of static CMOS circuits, an inverter circuit will be used to illustrate
the relatively large amount of power which a typical static CMOS circuit dissipates.
[0004] Fig. 1 of the accompanying drawings is a schematic circuit of two typical inverter
circuits which include a pair of P-channel devices Q26, Q30, and a pair of N-channel
devices Q28, Q32. The source electrodes of the P-channel devices are connected to
power supply (Vps) and the source electrodes of the N-channel devices are connected
to ground potential. As is the usual practice with static CMOS circuits without availability
of depletion NMOS devices, a P-channel FET device (Q26) is used as the load device.
To provide the loading function, the gate electrode of Q26 is connected to the ground
potential.
[0005] Current in the P-channel device can be expressed by the following simplified equation.
I = K (Vgate - Vs - Vt)², where
K is a function of parameters such as device width, length, gate oxide thickness,
etc.
Vgate is the gate voltage of the P-channel device
Vs is the source voltage of the device, the power supply Vps
Vt is the threshold voltage of the device, typically -1 volt
For the P-channel device shown (Vgate - Vs) = Vps
Vps is the power supply voltage typically 5 ± 10%
[0006] Variations in the process parameters cause K to vary by as much as +/- 60% between
its -3 sigma limit and its +3 sigma limit. Varying the power supply voltage Vps from
4.5 volts to 5.5 volts increases the current in the P-channel device by 50%. Combining
process variations and power supply tolerance, the current in the device can vary
more than 400% from worst case (-3 sigma limit) to best case (+3 sigma limit) The
-3 sigma and +3 sigma limits are statistical terms used to describe standard deviation
from a nominal value.
[0007] Circuit performance can improve by a factor of 4 between worst case conditions and
best case conditions. The designer usually has a worst case performance target for
the circuit being designed. The designer then must be able to accommodate the power
dissipation of the circuit at best case conditions. If there is a significant number
of static circuits on a VLSI chip, the power dissipation at best case conditions will
raise the chip's junction temperature, reduce its reliability and decrease the performance
of other circuits.
[0008] A straightforward approach for solving the power dissipation problem is to package
the chip in a ceramic casing. Such a casing usually has lower thermal resistance which
keeps the junction temperature lower. Ceramic packages have lower thermal resistance
than plastic packages but are more expensive and will make the component less competitive
in the market place. In addition, the increased current stresses the current density
limits of metal and contact structures further compounding the reliability issues.
[0009] The invention seeks to provide an on-chip circuit arrangement which biases static
CMOS circuits to reduce the maximum power which is dissipated by the circuits while
optimizing performance.
[0010] The invention provides a bias voltage generator for static circuits on a CMOS chip
comprising a first compensation circuit arrangement coupled to a power supply and
a control node, the arrangement serving to generate at the control node a control
voltage signal which compensates for tolerances in the power supply and device process
parameters; and a second compensation circuit arrangement coupled to the control node
and serving to adjust the control voltage signal at the control node to compensate
for device threshold voltage variations.
[0011] How the invention can be carried out will now be described by way of example, with
reference to Figs. 2 to 5 of the accompanying drawings, in which:-
Fig. 2 shows variations in circuit delay and P-channel current versus process parameters
and power supply tolerances;
Fig. 3 shows a VLSI chip including a bias voltage generator embodying the present
invention;
Fig. 4 shows a circuit schematic of an on-chip bias voltage generator embodying the
present invention; and
Fig. 5 shows a graph of the output voltage of the bias voltage generator as the current
in the load device varies between -3 sigma and +3 sigma.
[0012] Fig. 2 shows a graph of P-channel load current and circuit delay for load device
Q26 connected with its gate electrode connected to ground, as shown in the known inverter
circuit arrangement of Fig. 1. The graph is helpful in understanding the power/performance
problems which can be overcome by application of the present invention. As is shown
in Fig. 2, the statistical distribution or spread from worst case (-3 sigma limit)
to best case (+3 sigma limit) is plotted on the horizontal axis while the normalised
circuit delay or P-channel current ratio is plotted on the vertical axis. The variation
is caused by tolerances in the power supply voltage (Vps) and variation due to process
and/or device parameters. The curve which represents P-channel current has a positive
slope while the curve which represents circuit delay has a negative slope. A casual
review of the graph shows that as the device performance varies between -3 sigma and
+3 sigma the circuit performance varies by a ratio of 4.
[0013] The invention is concerned with a bias voltage generator to generate a bias voltage
for application to the gate or control electrode of the load device Q26 (Fig. 1) The
bias voltage generator compensates for variation due to process parameters and power
supply tolerances which reduces the current variation in the load device from 400%
to 40%.
[0014] Fig. 3 shows a VLSI chip 10 incorporating the present invention. The chip is exploded
or magnified beyond its normal size for purposes of explanation. In actuality, the
surface area of the chip is within millimeter range. Still referring to Fig. 3, the
chip comprises a substrate 12 upon which a plurality of circuit arrangements are integrated.
For brevity, only those circuit arrangements which are germane to explaining the present
invention are shown, it being understood that the chip is populated with other circuit
arrangements which are not shown in the sketch of Fig. 3. A plurality of static CMOS
circuit arrangements 14-1 through 14-N are integrated on substrate 12. The static
CMOS circuits may include PLAs, dotted circuits connected as bus structures, inverter
circuits, etc. Each of these circuit arrangements includes a pull-up device which
is biased by a bias voltage generator embodying the invention in order to minimize
power dissipation in those circuit arrangements.
[0015] Thus, the present invention provides an on-chip bias generator which outputs a voltage
Vreference (Vref) which is connected to the pullup devices (not shown) of each of
the static CMOS circuits. For example, if each of the static CMOS circuits 14-1 through
14-N includes an inverter circuit such as the prior art inverter circuit of Fig. 1,
the output Vreference from bias generator is connected to the gate electrode of the
P-channel enhancement mode device Q1. With this configuration the power performance
spread of the P-channel device is brought within acceptable range. Of course, a bias
voltage generator embodying the present invention may be used to bias other types
of static CMOS circuits.
[0016] Fig. 4 shows a circuit diagram of the voltage bias generator including a plurality
of compensating circuits which are coupled through an output circuit to provide an
output voltage which is applied to the gate electrodes of P-channel load devices.
The bias voltage compensates for variations due to the process parameters of device
channel length, channel width, and threshold voltage. The voltage is also adjusted
for variation in the power supply voltage (VDD).
[0017] Generally, compensation is achieved by the technique of ratioing nominal parameters
of two FET devices which are connected in series. A plurality of the series connected
devices are arranged in parallel to provide a nominal value for a desired output voltage.
The bias voltage generator includes a first compensation circuit arrangement 14, and
a second compensation circuit arrangement 16. The first compensation circuit arrangement
14 compensates for tolerance in the power supply (VDD) and variation in the device
parameters due to process variation. The second compensation circuit arrangement 16
adjusts for variation in device threshold voltage (Vt). The first and second compensation
circuit arrangements 14 and 16 are connected to node 2. Node 2 in turn is connected
by an output circuit arrangement 18 which produces V reference.
[0018] In the preferred embodiment of this invention, the output circuit arrangement 18
is an inverter circuit formed by devices Q6 and Q16. It should be noted that devices
with a diagonal such as Q16 and similar ones are P-channel enhancement mode devices
while devices with no diagonals such as Q6 and similar ones are N-channel enhancement
mode devices. Circuit arrangements 20 and 22 provide testability to the voltage generator.
[0019] Still referring to Fig. 4, the first compensation circuit means 14 includes N-channel
FET devices Q1 and Q2 which are connected in series between VDD and ground potential
and a P-channel FET device Q3 which is also connected to VDD. The first compensation
circuit 14 further includes a pair of N-channel FET devices Q4 and Q5 which are positioned
in parallel and connected to the drain electrode of Q3. The combination of Q3, Q4
and Q5 interconnects VDD to ground potential. Node 1 which is positioned between devices
Q1 and Q2 is connected to the gate electrode of Q4. In operation, devices Q1-Q5 bias
node 2 such that a desired bias voltage is achieved for most combinations of voltage
and parameters. The voltage at node 1 is designed to be about VDD/2. As the power
supply voltage is raised from 4.5 volts to 5.5 volts, the (Vgs-Vt) of Q4 increases
by approximately 40%. For devices Q3 and Q5 their Vgs-Vt increases by approximately
30% as VDD increases from 4.5 to 5.5 volts. To obtain the desired output voltage,
the voltage at node 2 must increase only 0.100 volt while VDD is raised from 4.5 to
5.5 volts. If only device Q5 were used, the voltage at node 2 would rise too much
as VDD increases. The effect of changes on the output voltage as VDD changes is shown
in Fig. 5. If only device Q4 were used, node 2 would decrease as VDD increases. By
choosing the correct device sizes for Q4 and Q5, the desired increase of 0.100 volt
at node 2 is achieved.
[0020] To compensate for variations in device parameters, the FETs are designed in large
and small widths and channel lengths. For best case parameters (highest device current)
the output voltage must be higher. Therefore, the voltage at node 2 must decrease
for best case (+3 sigma) parameters. The width and length of the device are a nominal
value +/- the tolerance.
DW = W nominal + tolerance (equation 1)
DL = L nominal + tolerance (equation 2)
where D represents Device
W represents Width
L represents Length
[0021] For a device with a large W nominal and L nominal, the current will be insensitive
to parameter variations (tolerance). The current in devices with small W nominal and
L nominal will increase significantly for best case parameters (+3 sigma) versus worst
case parameters (-3 sigma). Device Q3 (Fig. 4) has large geometries and devices Q4
and Q5 have small geometries so that the voltage at node 2 will decrease as parameters
vary from -3 sigma to +3 sigma. For example, the following set of parameters are suitable
values for devices Q3, Q4, Q5.
|
Width |
Length |
Q3 |
50 |
6 |
Q4 |
20 |
3 |
Q5 |
10 |
3 |
Tolerance = ±1 |
Hence, it can be seen that the width to length ratio (W/L) of device Q3 only varies
20% from nominal to +3σ while the W/L ration of Q5 varies by 65% from nominal to +3σ.
[0022] Still referring to Fig. 4, the second compensation circuit arrangement includes devices
Q7 through Q12. Each of a parallel pair of P-channel FET devices Q7 and Q10 is connected
in series with an individual one of N-channel devices Q8 and Q11. Devices Q7 and Q8
are connected at node 4 which in turn is connected to the gate electrode of device
Q9. The source electrode of device Q9 is connected to node 2 and the drain electrode
is connected to ground potential. Similarly, devices Q10 and Q11 are connected in
series at node 5 which is connected to the gate electrode of device Q12. Device Q12
is connected to node 2 and VDD. In operation, devices Q7-Q12 compensate for variations
in device threshold voltage (Vt). When the P-channel threshold voltage Vt is low (Vgs
- Vt large) the load device current (such as Q26), Fig. 1, will be too high. Current
in device Q7 (Fig. 4) will increase and the voltage at node 4 will rise turning on
device Q9. Current in device Q9 reduces the voltage at node 2 which reduces the current
capability of Q6. The bias voltage rises slightly, reducing the current in the load
device and in Q7. When the Vt of the P-channel devices is high (Vgs - Vt minimum)
current in the load devices is too low. Current in device Q10 decreases and the voltage
at node 5 drops which turns on Q12. Current in Q12 raises the voltage at node 2. A
higher voltage at node 2 increases the current at Q6, reducing the bias voltage and
increasing current in the load devices. Most combinations of power supply voltage
and process parameters of the voltage at nodes 4 and 5 are such that devices Q8 and
Q12 are off and do not have any effect on the output voltage.
[0023] The bias voltage generator can be deactivated by devices Q13-Q15 and Q17. P-channel
device Q13 is connected in series with N-channel device Q14 between VDD and ground
potential. When the +chip power is activated, devices Q13-Q15 cause the disabling
of all dc power dissipation within the bias generator and the Vref potential rises
to the full power supply potential. Vref = Vps reduces to zero all power dissipation
in the controlled elements with the design. Similarly, devices Q17 and Q15 deactivate
the output of the voltage generator when a signal is applied to the Degate input allowing
maximum dc power dissipation. The features associated with devices Q13-Q15, Q17 are
auxiliary to the bias generator function and provide the additional function of auto-override,
both for minimum and maximum power checking and testability.
1. A bias voltage generator for static circuits on a CMOS chip comprising
a first compensation circuit arrangement (14) coupled to a power supply (+VDD) and
a control node, the arrangement serving to generate at the control node a control
voltage signal which compensates for tolerances in the power supply and device process
parameters; and
a second compensation circuit arrangement (16) coupled to the control node and serving
to adjust the control voltage signal at the control node to compensate for device
threshold voltage variations.
2. A bias voltage generator as claimed in claim 1, wherein the first compensation
circuit arrangement includes a first circuit means (Q1, Q2) for generating a first
voltage, and
a second circuit means (Q3, Q4, Q5) responsive to the first voltage to generate the
control voltage at the control node.
3. A bias voltage generator as claimed in claim 2, wherein the first circuit means
includes a pair of N-channel FET devices (Q1, Q2) connected in series.
4. A bias voltage generator as claimed in claim 2 or claim 3, wherein the second circuit
means includes a P-channel FET device (Q3) coupled to a supply voltage source (+VDD),
and a pair of N-channel FET devices (Q4, Q5) coupled in parallel relative to one another
and each coupled in series relative to the P-channel device.
5. A bias voltage generator as claimed in any preceding claim, wherein the second
compensation circuit arrangement includes two pairs of series-connected FET devices
(Q7, Q8; Q10, Q11) positioned in a parallel configuration between a voltage supply
source (+VDD) and a ground potential; and two switching FET devices (Q9, Q12) connected
in series with each switching devices having its control electrode connected to an
individual one of the nodes between the devices in each pair of series-connected FET
devices.
6. A bias voltage generator as claimed in any preceding claim, further including an
inverting circuit means (18) coupled to the control node and serving to invert the
control voltage signal to the control node to produce the output voltage of the generator.
7. A bias voltage generator as claimed in any preceding claim, further including a
deactivating circuit (20) coupled to the first compensation circuit arrangement for
causing the latter to produce a chip-circuit deactivating control voltage at the control
node.
8. A bias voltage generator as claimed in any preceding claim, further including a
degating circuit means (22) coupled to the second compensation circuit arrangement
for deactivating the control node.
9. An integrated circuit chip incorporating static CMOS circuits and a bias voltage
generator as claimed in any preceding claim with its bias voltage output coupled to
the static CMOS circuits.